The Latest ADHD Neuroimaging Blockbuster: Stimulants Target Arousal, Not Attention

A massive neuroimaging study settles a debate and raises future questions

As the year comes to a close, I thought I would cover one more paper that garnered a few headlines. Earlier this month, a large scale neuroimaging paper relevant to how stimulants work was published in Cell. The study looked at ~12,000 children taking prescription stimulants using functional neuroimaging (brain scans) and concluded that stimulants primarily affect arousal and reward processing instead of attention.

I was happy reading this paper, but I was not particularly surprised. Since their discovery in the 1920s, stimulants had been used to treat narcolepsy until they were noticed to have paradoxical effects on a subset of hyperactive children. Charles Bradley’s 1937 observations documented that amphetamine “subdued” children with behavioral problems and improved their school performance. The original phenomenon was about wakefulness and behavioral management, not attention enhancement (attention is a complicated thing, more on how it is defined later in the piece).

But that did not stop the headlines: “Brain Scans Reveal a Surprise About ADHD Medications”. One outlet claimed “doctors’ thinking has been overturned”. The framing suggests doctors have been fundamentally wrong for decades about how these medications function. But were they, really?

No. The study in and of itself is not a huge surprise. It’s a very solid study and it needed to be done. However, on balance, it mainly resolves a messy fMRI literature where small studies reported conflicting results about whether stimulants modulate “attention networks.” The answer, with lots of data, is clear: they don’t. The effects indicate that the main impact is on arousal and reward processing.

But why did I decide to write about this now? The main motivation is that about 3.5 million children in the U.S. take stimulants for ADHD, and the number keeps climbing. Not everyone is benefiting from this intervention (and in fact, in some cases the risks outweigh the benefits) so understanding how these medications impact circuit function is important. It will allow us to match these medications to the right patients, and we are becoming better at measuring the circuit substrates that are predictive. In plain English: if stimulants work primarily through arousal and reward processing, helping with task engagement, then they should help people whose core problem is sustaining motivation and staying alert. They might not help much for people whose core problem is something else.

This piece will walk through what the study actually found, explain the clinical relevance, and make the case for why we need to think more carefully about what we’re treating when we prescribe stimulants. An important takeaway is that while ADHD may be a good starting point for a diagnostic label, we must move beyond it when making specific treatment decisions.

The Main Findings

This paper is a massive undertaking: Kay, Dosenbach, Fair and colleagues analyzed resting state fMRI data from the Adolescent Brain Cognitive Development Study, one of the largest pediatric neuroimaging datasets available.

Their main finding was straightforward: stimulants primarily changed the brain’s overall state. Sensory and motor systems showed reduced correlation within themselves and increased correlation with networks involved in salience and motivation. This is often interpreted to mean that the drugs changed the large-scale configuration of brain networks underlying task processing. These effects were the largest and most reliable effects observed. In contrast, networks typically labeled as attentional showed no measurable change, despite sufficient power to detect effects of the size previously reported in the literature.

The authors validated these findings in a separate drug trial with healthy adults. Participants were scanned extensively both on and off methylphenidate. The same pattern emerged, confirming that the effects were drug related rather than reflecting other differences between groups.

The most important result (and somewhat surprising) was related to sleep. Parent-reported sleep duration showed a brain connectivity pattern that closely matched the stimulant effect. These patterns also agreed with independent physiological measures of arousal, including EEG-based indices, respiratory variation during fMRI, and maps of norepinephrine transporter density from PET imaging.

This means that children who slept less showed brain signatures consistent with lower arousal. Among those same children, taking a stimulant eliminated that signature. Their brain measurement became indistinguishable from that of well-rested children. The same interaction appeared at the behavioral level: shorter sleep was associated with worse school performance, but this relationship was no longer present in children taking stimulants.

This makes sleep deprivation and stimulant treatment look like two manipulations of the same underlying arousal system.

The absence of effects in attention and control networks is also informative. Prior studies reporting stimulant effects in these systems relied on small samples and short scan durations. This study was powered to detect the previously reported effect sizes and did not find them. The negative result is therefore meaningful rather than inconclusive.

The fMRI literature on stimulants has been inconsistent for years. This study largely resolves that confusion by using scale, long scan durations, and data-driven network definitions. The most robust and replicable effects of stimulants in this study were in systems related to arousal and reward.

Attention in ADHD and Cognitive Psychology are Different

There are a few points worth clarifying at this stage. First, is how networks are referred to in different fields that measure brain substrates at various scales and, second, how cognitive constructs such as attention are defined across different disciplines.

Networks are very different things when measured by fMRI and when modeled as objects of computation. For example, the highlighted paper mentions the dorsal attention network (DAN), ventral attention network (VAN), and frontoparietal network (FPN) whose measurements do not appear to be reliably affected by stimulants. What are these networks? They are labels applied to clusters of brain regions that show correlated activity during rest. These labels came from observing which regions activate during tasks requiring attention or cognitive control. But these correlation patterns do not inform us about what these regions compute or exactly how they interact at a timescale relevant to thinking. Importantly, the finding that stimulants don’t change these resting-state patterns doesn’t in and of itself necessarily mean stimulants don’t affect attention. It means they don’t affect these particular correlations.

Now onto the second issue: “attention”. This word means different things in different contexts. In ADHD diagnostic criteria, “inattention” means not staying on task, getting distracted, forgetting instructions. These are problems with sustained engagement and persistence. By contrast, in cognitive psychology, attention refers to selective amplification of relevant information, like covertly allocating processing resources to a small patch of visual space (e.g. in a Posner cueing task; see figure below) or selectively increasing focus to one out of several conversation in a crowded room without actually moving your head (the cocktail party problem). That is not what people refer to when they are discussing “attention” in ADHD.

The imaging results therefore line up pretty nicely with the clinical meaning of attention in ADHD, which is about sustained engagement rather than selective processing. Whether stimulants actually help with selective attention in the cognitive psychology sense is not particularly clear because we lack sufficiently high powered studies looking at it from that perspective. That said, it is also unclear whether people with ADHD have selective attention deficits beyond those related to arousal and task engagement. Maybe a subset exists, but that hasn’t been demonstrated empirically as far as I’m aware.

Figure showing what a variant of the Posner attentional cueing task

The Unmet Need Beyond Stimulants

Let’s zoom out and big picture this a bit: the neuroimaging findings clarify what stimulants do, but they also reveal what they don’t do. The lack of effects in networks associated with attention raises a question: do stimulants actually improve cognitive control mechanisms, or do they just make it easier to stay engaged long enough for existing control processes to operate?

Cognitive control refers to the ability to flexibly organize behavior toward goals, manage interference from distractions, and maintain task rules. Controlling how attention is deployed is a cognitive control operation, and our lab has worked on this problem for a while. Many standard laboratory tests have been developed to measure cognitive control broadly. For example, the Stroop task (demo below) requires naming the color of a word while ignoring its meaning. Seeing “RED” printed in blue requires saying “blue” while suppressing the automatic response to read the word. Task switching, another cognitive control operation, measures the ability to rapidly alternate between different rules.

Studies of methylphenidate effects on these tasks show mixed results. Some report improvements in accuracy on Stroop interference or task switching, others find no effects. What improves consistently across studies is reaction time variability and sustained performance, both measures of arousal and engagement rather than cognitive control itself. This pattern confirms the theme: stimulants help by maintaining adequate arousal, but may not enhance cognitive control or its impact on attention.

This provisional conclusion is important because stimulants don’t work for everyone. Between 20 and 40 percent of patients don’t achieve adequate symptom control even with optimized dosing. The response rate improves to 75 to 90 percent if both methylphenidate and amphetamine formulations are tried, but that still leaves a meaningful fraction who don’t benefit. Some of this reflects poor adherence or intolerable side effects. But some may reflect genuine differences in what’s impaired. For patients whose barrier is arousal or motivation, stimulants may be a solution. For patients whose core problem is the cognitive control machinery itself, they may not.

There are also risks that constrain treatment.

Stimulants can trigger psychosis, particularly at higher doses. The overall risk is low, but it’s real and dose dependent. Amphetamines carry roughly twice the risk of methylphenidate. For many patients, such symptoms limit the utility of stimulants altogether.

What Now?

The neuroimaging result tells us what stimulants do. They change arousal state and reward processing. This explains why they work for certain patients with ADHD. Tasks feel more rewarding, effort is easier to sustain, focus persists longer.

But 20 to 40 percent of patients don’t achieve adequate symptom control even with optimized treatment. These patients present with the same behavioral symptoms, inattention, disorganization, difficulty sustaining effort, but the underlying problems are heterogenous, some of which we may not even understand.

The literature on cognitive control tasks suggests one possibility. Studies consistently show that stimulants improve reaction time variability and sustained performance, both reflecting engagement and arousal. But effects on accuracy are mixed. Some studies find improvements in Stroop interference or task switching, others find nothing. This inconsistency supports the notion of heterogeneity.

Schizophrenia provides the clearest boundary case, where boosting arousal and salience fails to improve disorganized cognition. Stimulants are not used routinely in schizophrenia because they tend to worsen positive symptoms, particularly in patients with poor symptom control.

Some ADHD patients may have similar computational perturbations related to cognitive control but we currently lack ways to identify them prospectively and limited interventions that target control processes directly (no FDA approved treatments specifically for this problem as far as I’m aware).

The path forward requires systematic research. We need better measurement tools that can distinguish putative ADHD subtypes and help triage people prospectively. This work is poised to identify circuit substrates both as biomarkers and treatment targets. Then we need clinical trials testing whether targeting specific mechanisms improves outcomes compared to the current trial-and-error approach. The will for this new chapter of psychiatry is there. What’s needed now is the next generation of researchers to take on the challenge and the investments to support them.

---

If you enjoyed this piece, please feel free to share it with your community. I write about topics related to neuroscience, mental health and their overlap. You can subscribe to join the conversation and stay updated.